Problem/Discussion Set for “Chaos and ...

advertisement
Harvard-MIT Division of Health Sciences and Technology
HST.750: Modeling Issues in Speech and Hearing, Spring 2006
Course Directors: Dr. Christopher Shera and Dr. Jennifer Melcher
HST–750
M ODELING I SSUES IN S PEECH AND H EARING
Problem/Discussion Set for “Chaos and the Limits to Prediction”
1. Sensitive Dependence on Initial Conditions. Equations 25–27 in Lorenz (1963) define a system
of coupled nonlinear equations known as the Lorenz equations.1 They can be written in the form
xP D f.x/, where x. / D .X. /; Y . /; Z. //.
(a) Using numerical integration (e.g., the function ode23 in Matlab), integrate the Lorenz equations
over the time interval 2 Œ0; 100�. Start your integration at the initial point x.0/ D .20; 5; 5/.
Follow Lorenz and use the parameters values D 10, b D 8=3, r D 28, and a time increment
� D 0:01. Plot the trajectory x. / in three dimensions (e.g., using Matlab’s plot3).2
(b) Now integrate the Lorenz equations starting at 50 or more different nearby initial points chosen
by selecting points xi .0/ at random from within a cube �X�Y�Z in phase space with dimen­
sions �X D �Y D �Z D 10 3 . Center your cube at the point x D .20; 5; 5/. At each
of the six times D f6; 9; 12; 15; 18; 21g plot the position of each of the 50 trajectories. Use a
separate graph for each of the six time points, and, to facilitate comparisons, give all your plots
the same scale. Use Lorenz’s Eq. 37 to compute the approximate fractional change in volume
of your cube that occurs over the total time interval D Œ0; 21�. What happens to your cube of
initial conditions?
2. Lyapunov Exponent for the Tent Map. The extent to which systems show sensitive dependence
on initial conditions can be quantified by measuring (or calculating) what is known as the Lyapunov
exponent. Consider two nearby initial conditions (or point) x0 and x0 C ı0 , where the initial separa­
tion jı0j is small. How does the distance between the trajectories starting at these two initial points
evolve under the operation of the map xnC1 D f .xn /? To answer this question quantitatively, let
the separation after n iterations be denoted ın. If the separation jın j varies exponentially with n, then
jın j ' jı0je n , where is the Lyapunov exponent. Lyapunov exponents greater than zero ( > 0)
indicate that nearby trajectories diverge exponentially fast (i.e., chaotic behavior).
(a) Show that
ˇ
1 ˇˇ n 0
ln .f / .x0 /ˇ :
n
Hint: Solve for using the fact that ın D f n .x0 C ı0/ f n .x0/ and take the limit ı0 ! 0.
In these expressions, the superscripts on the function f represents functional composition (or
iteration). For example, f 2.x/ f .f .x//.
(b) Use the chain rule to show that
n 1
ˇ
1 X ˇˇ 0
ln f .xi /ˇ :
n
iD0
We now define the Lyapunov exponent by the limit n ! 1:
)
( n 1
ˇ
1 X ˇˇ 0
lim
ln f .xi /ˇ :
n!1 n
iD0
(c) One can generalize Lorenz’s Eq. 35 by defining a “tent map” with the equations
(
for Mn < 12
rMn
MnC1 D f .Mn / D
r .1 Mn / for Mn > 12
1 To
see them in action, try the lorenz demo in Matlab.
credit: Compute the amount of computation time per time step required to solve the equations and compare with the value reported by
Lorenz (“one second per iteration, aside from output time”).
2 Extra
HST–750
M ODELING I SSUES IN S PEECH AND H EARING
for 0 r 2 and 0 M 1. Plot f .Mn /, f 2 .Mn /, and f 3 .Mn / assuming, as Lorenz did,
that r D 2. Use your plots to verify the existence of “a single one­phase, a single two­phase, and
two three­phase sequences, namely
2=3;
2=5;
2=7;
2=9;
:::;
4=5; : : : ; 4=7; 6=7; : : : ; 4=9; 8=9; : : : :”
(d) Calculate for the tent map. How does the long­term behavior of the system depend on the value
of r ?
3. Orbit Diagram for the Quadratic Map. Following Feigenbaum, define the iterated quadratic (or
logistic) map as
xkC1 D 4xk .1 xk / :
(a) Choose a random number x0 in the interval Œ0; 1�. Plot the resulting sequence fxk g for k D
0; 1; : : : ; 100 for each of the following values of r 4: r D f2:8; 3:3; 3:5; 3:55; 3:9g. Repeat
each for several different random choices for x0 .
(b) Now construct a plot that shows the long term behavior of xk for many values of r simulta­
neously. To do this, first chose an initial value of r and some initial starting point, x0. Then
generate the resulting sequence fxk g, iterating until the transients, which depend on the choice
of x0 , have died down and the system has settled into its long term behavior (typically, iterating
until k D 300 or so should be sufficient). Now generate many more points xk (i.e., generate and
store the values x301 to x600) and plot the values of xk>300 versus the value of r that you chose
above. (In the xy­plane your plot will have points at 300 or so values of y—not necessarily all
different—for every point, r , along the x­axis.) Now move to a nearby value of r and repeat the
process. In this way, you’ll build up a plot of xk>300 versus r . Vary r over the range r D Œ2:9; 4�
with a resolution �r of at least 0.002.
(c) Use your plot to estimate, as accurately as you can, the values rn at which a cycle of period 2n
first appears (you may want to zoom in and compute certain regions of your plot with greater
resolution). For example, you should find that a cycle of length 2 appears at r1 D 3. Use your
values of rn to estimate the ratios
rn rn 1
ın D
rnC1 rn
for n D 2; 3, and 4. Compare your answers with the limiting, universal value
rn rn 1
D 4:66920160910299067 : : :
n!1 rnC1
rn
ı D lim
known as Feigenbaum’s constant.3
4. Implications.
(a) In an interview on National Public Radio, the physicist Peter Carruthers said that the phenomenon
of chaos (i.e., sensitive dependence on initial conditions) “undermines the whole basis of sci­
ence.” What do you suppose he was talking about? Explain why you agree or disagree. If you
disagree, what does chaos do to science?
(b) In the talk in which he coined the term “butterfly effect,” Lorenz suggested that the flap of a
butterfly’s wings in Brazil can set off a tornado in Texas. Can it? Why or why not?
3 In
1999, Feigenbaum’s constant was computed to 1018 decimal places. (The record for is now over 1.24 trillion decimal places.)
Download